view libvo/vo_svga.c @ 9660:c2d23e02522b

improvements to detc filter: -> use of 8x8 blocks rather than 16x16 to better localize the search for interlacing. this helps detect interlacing in very small motions, e.g. mouths in anime. -> removed some redundant conditions in the logic -> looser condition for detecting lacing and more forgiving of slight mismatches between fields from the two telecine frames to make up for quantization noise in low quality encodes. this code is still mostly experimental but probably better than the old version, so maybe it should be backported to 0.90...?

/*

  Video driver for SVGAlib 
  by Zoltan Mark Vician <[e-mail]>
  Code started: Mon Apr  1 23:25:47 2001
  Some changes by Matan Ziv-Av <[e-mail]>
  Compleat rewrite by Ivan Kalvachev 19 Mar 2003:

Wrangings:
 -  1bpp doesn't work right for me with '-double' and svgalib 1.4.3, 
    but works OK with svgalib 1.9.17
 -  The HW acceleration is not tested - svgalibs supports few chipsets, 
    and i don't have any of them. If it works for you then let me know. 
    I will remove this warning after confirm its status.
 -  retrace sync works only in doublebuffer mode.
 -  the retrace sync may slow down decoding a lot - mplayer is blocked while
    waiting for retrace
 -  denoise3d fails to find common colorspace, use -vf denoise3d,scale
   
TODO:
 - let choose_best_mode take aspect into account
 - set palette from mpi->palette or mpi->plane[1]
 - let OSD draw in black bars - need some OSD changes
 - Make nicer CONFIG parsing
 - change video mode logical width to match img->stride[0] - for HW only 
*/

#include <stdio.h>
#include <stdlib.h>
#include <string.h>

#include <vga.h>

#include <limits.h>

#include "config.h"
#include "video_out.h"
#include "video_out_internal.h"
#include "fastmemcpy.h"
#include "osdep/getch2.h"
#ifdef CONFIG_VIDIX
#include "vosub_vidix.h"
#endif

#include "sub.h"

#include "../mp_msg.h"
//#include "../mp_image.h"

#include <assert.h>

//silence warnings, probably it have to go in some global header
#define UNUSED(x) ((void)(x)) 

extern int vo_doublebuffering;
extern int vo_directrendering;
extern int vo_dbpp;
extern int verbose;

static uint32_t query_format(uint32_t format);
static void draw_alpha(int x0, int y0, int w, int h, unsigned char *src,
                       unsigned char *srca, int stride);
static uint32_t get_image(mp_image_t *mpi);

#define MAXPAGES 16

#define CAP_ACCEL_CLEAR 8
#define CAP_ACCEL_PUTIMAGE 4
#define CAP_ACCEL_BACKGR 2
#define CAP_LINEAR 1

static uint8_t zerobuf[8192];//used when clear screen with vga_draw

static int squarepix;
static int force_vm=0;
static int force_native=0;
static int sync_flip=0;
static int cpage=0, max_pages;

static vga_modeinfo * modeinfo;
static int mode_stride; //keep it in case of vga_setlogicalwidth
static int stride_granularity; //not yet used
static int mode_bpp;
static int mode_capabilities;

static int image_width,image_height; // used by OSD
static uint32_t x_pos, y_pos;

static struct {
  int yoffset;//y position of the page
  int doffset;//display start of the page
  void * vbase;//memory start address of the page
}PageStore[MAXPAGES];

static vo_info_t info = {
    "SVGAlib",
    "svga",
    "Ivan Kalvachev <[e-mail]>",
    ""
};

#ifdef CONFIG_VIDIX
static char vidix_name[32] = "";
#endif

LIBVO_EXTERN(svga)

/*
probably this should be in separate pages.c file so 
all vo_drivers can use it, do it if you like it.
TODO
direct render with IP buffering support - for vo_tdfx_vid
*/

#define PAGES_MAX MAXPAGES
#define Page_Empty 0
#define Page_Durty 1

static int page_locks[PAGES_MAX];
static int pagecurrent;
static int pagetoshow;
static int pagesmax;

static void page_init(int max){
int i;
  for(i=0;i<max;i++){
    page_locks[i]=0;
  }
  pagetoshow=pagecurrent=0;
  pagesmax=max;
}
//internal use function
//return number of 1'st free page or -1 if no free one
static int page_find_free(){
int i;
  for(i=0;i<pagesmax;i++)
  if(page_locks[i]==Page_Empty) return i;
  return -1;
}

//return the number of page we may draw directly of -1 if no such
static int page_get_image(){
int pg;
  pg=page_find_free();
  if(pg>=0) page_locks[pg]=Page_Durty;
  return pg;
}

//return the number of page we should draw into.
static int page_draw_image(){
int pg;
  pg=page_find_free();
  if(pg<0) pg=pagecurrent;
  page_locks[pg]=Page_Durty;
  pagetoshow=pg;
  return pg;
}

static void page_draw_gotten_image(int pg){
assert((pg>=0)&&(pg<PAGES_MAX));
  pagetoshow=pg;
}

//return the number of the page to show
static int page_flip_page(){
  page_locks[pagecurrent]=Page_Empty;
  pagecurrent=pagetoshow;
assert(pagecurrent>=0);
  //movequeue;
  page_locks[pagecurrent]=Page_Durty;
  return pagecurrent;
}
//------------ END OF PAGE CODE -------------

static uint32_t preinit(const char *arg)
{
int i;
char s[64];

  getch2_disable();
  memset(zerobuf,0,sizeof(zerobuf));
  force_vm=force_native=squarepix=0;
  
  if(arg)while(*arg) {
#ifdef CONFIG_VIDIX  
    if(memcmp(arg,"vidix",5)==0) {
      i=6;
      while(arg[i] && arg[i]!=':') i++;
      strncpy(vidix_name, arg+6, i-6);
      vidix_name[i-5]=0;
      if(arg[i]==':')i++;
      arg+=i;
      vidix_preinit(vidix_name, &video_out_svga);
    }
#endif
    if(!strncmp(arg,"sq",2)) {
      squarepix=1;
      arg+=2;
      if( *arg == ':' ) arg++;
    }
    
    if(!strncmp(arg,"native",6)) {
      force_native=1;
      arg+=6;
      if( *arg == ':' ) arg++;
    }
 
    if(!strncmp(arg,"retrace",7)) {
      sync_flip=1;
      arg+=7;
      if( *arg == ':' ) arg++;
    }
    
    if(*arg) {
      i=0;
      while(arg[i] && arg[i]!=':')i++;
      strncpy(s, arg, i);
      s[i]=0;
      arg+=i;
      if(*arg==':')arg++;
      i=vga_getmodenumber(s);
      if(i>0) {
        force_vm = i;
        if(verbose)
        printf("vo_svga: Forcing mode %i\n",force_vm);
      }
    }
  }
  
  vga_init();
  return 0;
}

static void svga_clear_box(int x,int y,int w,int h){
uint8_t * rgbplane;
int i;

  if (mode_capabilities&CAP_ACCEL_CLEAR){
    if(verbose > 2)
      printf("vo_svga: clearing box %d,%d - %d,%d with HW acceleration\n",
             x,y,w,h);
    if(mode_capabilities&CAP_ACCEL_BACKGR)  
      vga_accel(ACCEL_SYNC);
    vga_accel(ACCEL_SETFGCOLOR,0);//black
    vga_accel(ACCEL_FILLBOX,x,y,w,h);
    return;
  }
  if (mode_capabilities & CAP_LINEAR){
    if(verbose > 2)
      printf("vo_svga: clearing box %d,%d - %d,%d with memset\n",x,y,w,h);
    rgbplane=PageStore[0].vbase + (y*mode_stride) + (x*modeinfo->bytesperpixel);
    for(i=0;i<h;i++){
//i'm afraid that memcpy is better optimized than memset;)
      memcpy(rgbplane,zerobuf,w*modeinfo->bytesperpixel);
//    memset(rgbplane,0,w*modeinfo->bytesperpixel);
      rgbplane+=mode_stride;
    }
    return;
  }
  //native
  if(verbose > 2)
    printf("vo_svga: clearing box %d,%d - %d,%d with native draw \n",x,y,w,h);
  if(modeinfo->bytesperpixel!=0) w*=modeinfo->bytesperpixel;
  for(i=0;i<h;i++){
    vga_drawscansegment(zerobuf,x,y+i,w);
  }
};

static uint32_t svga_draw_image(mp_image_t *mpi){
int i,x,y,w,h;
int stride;
uint8_t *rgbplane, *base;
int bytesperline;

  if(mpi->flags & MP_IMGFLAG_DIRECT){
    if(verbose > 2)
      printf("vo_svga: drawing direct rendered surface\n");
    cpage=(uint32_t)mpi->priv;
    page_draw_gotten_image(cpage);
    return VO_TRUE; //it's already done
  }
//  if (mpi->flags&MP_IMGFLAGS_DRAWBACK) 
//  return VO_TRUE;//direct render method 2
  cpage=page_draw_image();
// these variables are used in loops  
  x = mpi->x;
  y = mpi->y;
  w = mpi->w;
  h = mpi->h;
  stride = mpi->stride[0]; 
  rgbplane = mpi->planes[0] + y*stride + (x*mpi->bpp)/8;
  x+=x_pos;//center
  y+=y_pos;
  
  if(mpi->bpp >= 8){//for modes<8 use only native
    if( (mode_capabilities&CAP_ACCEL_PUTIMAGE) && (x==0) && (w==mpi->width) &&
        (stride == mode_stride) ){ //only monolite image can be accelerated
      w=(stride*8)/mpi->bpp;//we transfer pixels in the stride so the source
//ACCELERATE
      if(verbose>2) 
        printf("vo_svga: using HW PutImage (x=%d,y=%d,w=%d,h=%d)\n",x,y,w,h);
      if(mode_capabilities & CAP_ACCEL_BACKGR)
        vga_accel(ACCEL_SYNC);

      vga_accel(ACCEL_PUTIMAGE,x,y+PageStore[cpage].yoffset,w,h,rgbplane);
      return VO_TRUE;
    }
  
    if( mode_capabilities&CAP_LINEAR){
//DIRECT  
      if(verbose>2) 
        printf("vo_svga: using Direct memcpy (x=%d,y=%d,w=%d,h=%d)\n",x,y,w,h);
      bytesperline=(w*mpi->bpp)/8;
      base=PageStore[cpage].vbase + (y*mode_stride) + (x*mpi->bpp)/8;     

      for(i=0;i<h;i++){
        mem2agpcpy(base,rgbplane,bytesperline);
        base+=mode_stride;
        rgbplane+=stride;
      }
      return VO_TRUE;
    }  
  }//(modebpp>=8
  

//NATIVE
  {
  int length;
    length=(w*mpi->bpp)/8;
  //one byte per pixel! svgalib innovation
    if(mpi->imgfmt==IMGFMT_RG4B || mpi->imgfmt==IMGFMT_BG4B) length=w;
  
    if(verbose>2) 
      printf("vo_svga: using Native vga_draw(x=%d,y=%d,w=%d,h=%d)\n",x,y,w,h);
    y+=PageStore[cpage].yoffset;//y position of the page beggining
    for(i=0;i<h;i++){
      vga_drawscansegment(rgbplane,x,y+i,length);
      rgbplane+=stride;
    }
  }
  return VO_TRUE;
}

int bpp_from_vminfo(vga_modeinfo *vminfo){
  switch(vminfo->colors){
    case 2: return 1;
    case 16: return 4;
    case 256: return 8;
    case 32768: return 15;
    case 65536: return 16;
    case 1<<24: return 8*vminfo->bytesperpixel;
  } 
  return 0;
}

int find_best_svga_mode(int req_w,int req_h, int req_bpp){
 int badness,prev_badness;
 int bestmode,lastmode;
 int i;
 vga_modeinfo *vminfo;
//int best aspect mode // best linear mode // best normal mode (no modeX)

  prev_badness = 0;//take care of special case below
  bestmode = 0; //0 is the TEXT mode
  lastmode = vga_lastmodenumber();
  for(i=1;i<lastmode;i++){
    vminfo = vga_getmodeinfo(i);
    if( vminfo == NULL ) continue;
    if( vga_hasmode(i) == 0 ) continue;
    if( req_bpp != bpp_from_vminfo(vminfo) )continue;
    if( (vminfo->width < req_w) || (vminfo->height < req_h) ) continue;
    badness=(vminfo->width * vminfo->height) - (req_h * req_w);
    //put here aspect calculations
    if(squarepix) 
      if( vminfo->width*3 != vminfo->height*4 ) continue;

    if( bestmode==0 || prev_badness >= badness ){//modeX etc...
      prev_badness=badness;
      bestmode=i;
      if(verbose>3)
        printf("vo_svga: found good mode %d with badness %d\n",i,badness);
    }
  }
  return bestmode;
}

static uint32_t control(uint32_t request, void *data, ...)
{
  switch (request) {
    case VOCTRL_QUERY_FORMAT:
      return query_format(*((uint32_t*)data));
    case VOCTRL_DRAW_IMAGE:
      return svga_draw_image( (mp_image_t *)data);
    case VOCTRL_GET_IMAGE:
      return get_image(data);
  }
  return VO_NOTIMPL;
}

//
// This function is called to init the video driver for specific mode
//
static uint32_t config(uint32_t width, uint32_t height, uint32_t d_width,
                       uint32_t d_height, uint32_t flags, char *title, 
                       uint32_t format) {
  int32_t req_w = width;// (d_width > 0 ? d_width : width);
  int32_t req_h = height;// (d_height > 0 ? d_height : height);
  uint16_t vid_mode = 0;
  int32_t req_bpp;
  
  uint32_t accflags;
  if(verbose)
    printf("vo_svga: config(%i, %i, %i, %i, %08x, %s, %08x)\n", width, height, 
           d_width, d_height, flags, title, format);
//Only RGB modes supported
  if (!IMGFMT_IS_RGB(format) && !IMGFMT_IS_BGR(format)) {assert(0);return -1;} 
  req_bpp = IMGFMT_BGR_DEPTH(format);
    
  if( vo_dbpp!=0 && vo_dbpp!=req_bpp) {assert(0);return-1;}
    
  if(!force_vm) {
    if (verbose) {
      printf("vo_svga: Looking for the best resolution...\n");
      printf("vo_svga: req_w: %d, req_h: %d, bpp: %d\n",req_w,req_h,req_bpp);
    }
    vid_mode=find_best_svga_mode(req_w,req_h,req_bpp);
    if(vid_mode==0) 
      return 1;
    modeinfo=vga_getmodeinfo(vid_mode);
  }else{//force_vm
    vid_mode=force_vm;
    if(vga_hasmode(vid_mode) != 0){
      printf("vo_svga: forced vid_mode %d (%s) not available\n",
             vid_mode,vga_getmodename(vid_mode));
      return 1; //error;
    }
    modeinfo=vga_getmodeinfo(vid_mode);
    if( (modeinfo->width < req_w) || (modeinfo->height < req_h) ){
      printf("vo_svga: forced vid_mode %d (%s) too small\n",
             vid_mode,vga_getmodename(vid_mode));
      return 1;
    }
  }
  mode_bpp=bpp_from_vminfo(modeinfo);
     
  printf("vo_svga: vid_mode: %d, %dx%d %dbpp\n",
         vid_mode,modeinfo->width,modeinfo->height,mode_bpp);
  
  if (vga_setmode(vid_mode) == -1) {
    printf("vo_svga: vga_setmode(%d) failed.\n",vid_mode);
    uninit();
    return 1; // error
  }
  /* set 332 palette for 8 bpp */
  if(mode_bpp==8){
    int i;
    for(i=0; i<256; i++)
      vga_setpalette(i, ((i>>5)&7)*9, ((i>>2)&7)*9, (i&3)*21);
  }
  /* set 121 palette for 4 bpp */
  else if(mode_bpp==4){
    int i;
    for(i=0; i<16; i++)
      vga_setpalette(i, ((i>>3)&1)*63, ((i>>1)&3)*21, (i&1)*63);
  } 
  //if we change the logical width, we should know the granularity
  stride_granularity=8;//according to man vga_logicalwidth
  if(modeinfo->flags & EXT_INFO_AVAILABLE){
     stride_granularity=modeinfo->linewidth_unit;
  }
  //look for hardware acceleration
  mode_capabilities=0;//NATIVE;
  if(!force_native){//if we want to use only native drawers
    if(modeinfo->flags & HAVE_EXT_SET){//support for hwaccel interface
      accflags=vga_ext_set(VGA_EXT_AVAILABLE,VGA_AVAIL_ACCEL);
      if(accflags & ACCELFLAG_FILLBOX) // clear screen
        mode_capabilities|=CAP_ACCEL_CLEAR;
      if(accflags & ACCELFLAG_PUTIMAGE)//support for mem->vid transfer
        mode_capabilities|=CAP_ACCEL_PUTIMAGE;
      if((accflags & ACCELFLAG_SETMODE) && (accflags & ACCELFLAG_SYNC)){
        vga_accel(ACCEL_SETMODE,BLITS_IN_BACKGROUND);
        mode_capabilities|=CAP_ACCEL_BACKGR;//can draw in backgraund
      }
    }
    if(modeinfo->flags & IS_LINEAR){ 
      mode_capabilities|=CAP_LINEAR; //don't use bank & vga_draw
    }
    else{
      if(modeinfo->flags & CAPABLE_LINEAR){
        int vid_mem_size;
        vid_mem_size = vga_setlinearaddressing();
        if(vid_mem_size != -1){
          modeinfo=vga_getmodeinfo(vid_mode);//sometimes they change parameters
          mode_capabilities|=CAP_LINEAR;
        }
      }
    }
  }//fi force native
  if(mode_capabilities&CAP_LINEAR){
    printf("vo_svga: video mode is linear and memcpy could be used for image transfer\n");
  }
  if(mode_capabilities&CAP_ACCEL_PUTIMAGE){
    printf("vo_svga: video mode have hardware acceleration and put_image could be used\n");
    printf("vo_svga: If it works for you i would like to know \nvo_svga: (send log with `mplayer test.avi -v -v -v -v &> svga.log`). Thx\n");
  }
  
//here is the place to handle strides for accel_ modes;
  mode_stride=modeinfo->linewidth;
//we may try to set a bigger stride for video mode that will match the mpi->stride, 
//this way we will transfer more data, but HW put_image can do it in backgraund!

//now lets see how many pages we can use  
  max_pages = modeinfo->maxpixels/(modeinfo->height * modeinfo->width);
  if(max_pages > MAXPAGES) max_pages = MAXPAGES;
  if(!vo_doublebuffering) max_pages=1;
//fill PageStore structs
  {
  int i;
  uint8_t * GRAPH_MEM;
  int dof;
    GRAPH_MEM=vga_getgraphmem();
    for(i=0;i<max_pages;i++){
    //calculate display offset
      dof = i * modeinfo->height * modeinfo->width;
      if(modeinfo->bytesperpixel != 0) dof*=modeinfo->bytesperpixel;
    //check video chip limitations
      if( dof != (dof & modeinfo->startaddressrange) ){
        max_pages=i;//page 0 will never come here
        break;
      }
      PageStore[i].yoffset = i * modeinfo->height;//starting y offset
      PageStore[i].vbase = GRAPH_MEM + i*modeinfo->height*mode_stride; //memory base address
      PageStore[i].doffset = dof; //display offset    
    }
  }
  assert(max_pages>0);
  printf("vo_svga: video mode have %d page(s)\n",max_pages);
  page_init(max_pages);
  //15bpp
  if(modeinfo->bytesperpixel!=0)
    vga_claimvideomemory(max_pages * modeinfo->height * modeinfo->width * modeinfo->bytesperpixel);
  else
    vga_claimvideomemory(max_pages * modeinfo->height * modeinfo->width * mode_bpp / 8);

  svga_clear_box(0,0,modeinfo->width,modeinfo->height * max_pages);

  image_height=req_h;
  image_width=req_w;
  x_pos = (modeinfo->width  - req_w) / 2;
  y_pos = (modeinfo->height - req_h) / 2;
  x_pos &= ~(15); //align x offset position to 16 pixels
  printf("vo_svga: centering image. start at (%d,%d)\n",x_pos,y_pos);

#ifdef CONFIG_VIDIX

  if(vidix_name[0]){ 
    vidix_init(width, height, x_pos, y_pos, modeinfo->width, modeinfo->height, 
        format, mode_bpp, modeinfo->width,modeinfo->height);
    printf("vo_svga: Using VIDIX. w=%i h=%i  mw=%i mh=%i\n",width,height,
           modeinfo->width,modeinfo->height);
    vidix_start();
  }
#endif    

  vga_setdisplaystart(0);
  return (0);
}

static uint32_t draw_slice(uint8_t *image[],int stride[],
               int w, int h, int x, int y) {
assert(0);
UNUSED(image);UNUSED(stride);
UNUSED(w);UNUSED(h);
UNUSED(x);UNUSED(y);

  return VO_ERROR;//this is yv12 only -> vf_scale should do all transforms
}

static uint32_t draw_frame(uint8_t *src[]) {
assert(0);
UNUSED(src);
  return VO_ERROR;//this one should not be called
}

static void draw_osd(void)
{
// for now draw only over the image
//  vo_draw_text(modeinfo->width, modeinfo->height, draw_alpha);//black bar OSD
  vo_draw_text(image_width, image_height, draw_alpha);
}

static void flip_page(void) {
static int oldpage=-1;
int page;
  page = page_flip_page();
  if(verbose > 2)
    printf("vo_svga: viewing page %d\n",page);
  if(sync_flip && oldpage!=page){
    vga_waitretrace();
    printf("vo_svga:vga_waitretraced\n");
  }
  vga_setdisplaystart(PageStore[cpage].doffset);
  oldpage=page;
}

static void check_events(void) {
}

static void uninit(void) {

#ifdef CONFIG_VIDIX
  if(vidix_name[0])vidix_term();
#endif
  vga_setmode(TEXT);
}

/* --------------------------------------------------------------------- */
static uint32_t query_format(uint32_t format) {
int32_t req_bpp,flags;
int i,lastmode;
vga_modeinfo * vminfo;

  if (verbose >3)
    printf("vo_svga: query_format=%X \n",format);
//only RGB modes supported
  if( (!IMGFMT_IS_RGB(format)) && (!IMGFMT_IS_BGR(format)) ) return 0; 

// Reject different endian
#ifdef WORDS_BIGENDIAN
  if (IMGFMT_IS_BGR(format)) return 0;
#else
  if (IMGFMT_IS_RGB(format)) return 0;
#endif

  //svgalib supports only BG4B! if we want BGR4 we have to emulate it (sw)
  if( format==IMGFMT_BGR4 || format==IMGFMT_RGB4) return 0;
  req_bpp = IMGFMT_RGB_DEPTH(format);
  if( vo_dbpp>0 && vo_dbpp!=req_bpp ) return 0; //support -bpp options
//scan all modes
  lastmode = vga_lastmodenumber();
  for(i=1;i<lastmode;i++){
    vminfo = vga_getmodeinfo(i);
    if( vminfo == NULL ) continue;
    if( vga_hasmode(i) == 0 ) continue;
    if( req_bpp != bpp_from_vminfo(vminfo) ) continue;
    if( (force_vm > 0) && (force_vm != i) )  continue;//quick hack
    flags = VFCAP_CSP_SUPPORTED|
      VFCAP_CSP_SUPPORTED_BY_HW|
      VFCAP_ACCEPT_STRIDE|
      0;
    if(req_bpp>8) flags|=VFCAP_OSD;
    return flags;
  }
  return 0;
}

static void draw_alpha(int x0, int y0, int w, int h, unsigned char *src,
                       unsigned char *srca, int stride) {
  char* base;
  int bytelen;

  if(verbose>2)
    printf("vo_svga: draw_alpha(x0=%d,y0=%d,w=%d,h=%d,src=%p,srca=%p,stride=%d\n",
           x0,y0,w,h,src,srca,stride);
  x0+=x_pos;// in case of drawing over image
  y0+=y_pos;
  
  //only modes with bytesperpixel>0 can draw OSD
  if(modeinfo->bytesperpixel==0) return;
  if(!(mode_capabilities&CAP_LINEAR)) return;//force_native will remove OSD
  if(verbose>3)
    printf("vo_svga: OSD draw in page %d",cpage);
  base=PageStore[cpage].vbase + y0*mode_stride + x0*modeinfo->bytesperpixel;
  bytelen = modeinfo->width * modeinfo->bytesperpixel;   
  switch (mode_bpp) {
    case 32: 
      vo_draw_alpha_rgb32(w, h, src, srca, stride, base, bytelen);
      break;
    case 24: 
      vo_draw_alpha_rgb24(w, h, src, srca, stride, base, bytelen);
      break;
    case 16:
      vo_draw_alpha_rgb16(w, h, src, srca, stride, base, bytelen);
      break;
    case 15:
      vo_draw_alpha_rgb15(w, h, src, srca, stride, base, bytelen);
      break;
  }
}

static uint32_t get_image(mp_image_t *mpi){
int page;

  if(!IMGFMT_IS_BGR(mpi->imgfmt) && !IMGFMT_IS_RGB(mpi->imgfmt) ){ 
    assert(0);//should never happen
    return(VO_FALSE);
  }
  
  if (
   ( (mpi->type != MP_IMGTYPE_STATIC) && (mpi->type != MP_IMGTYPE_TEMP)) ||
   (mpi->flags & MP_IMGFLAG_PLANAR) ||
   (mpi->flags & MP_IMGFLAG_YUV)
      )
    return(VO_FALSE);

//reading from video memory is horribly slow
  if( !(mpi->flags & MP_IMGFLAG_READABLE) && vo_directrendering &&
       (mode_capabilities & CAP_LINEAR) ){
    page=page_get_image();
    if(page >= 0){
      mpi->flags |= MP_IMGFLAG_DIRECT;
      mpi->stride[0] = mode_stride;
      mpi->planes[0] = PageStore[page].vbase + 
             y_pos*mode_stride + (x_pos*mpi->bpp)/8;
      (int)mpi->priv=page;
      if(verbose>2)
        printf("vo_svga: direct render allocated! page=%d\n",page);
      return(VO_TRUE);
    }
  }

  return(VO_FALSE);
}